slides - ETS home

Transcription

slides - ETS home
Innovation & Wealth Creation
from Technology
(With a Test Flavour)
Sir Robin Saxby
Chairman ARM Holdings plc
Visiting Professor University of Liverpool
Deputy President IET
1
H.M.Government & Innovation
May 03 DTI: ‘Competitiveness Study’,
shows UK Industry must become more
Innovative to remain competitive
Dec 03 DTI: ‘Innovation Report’, Pref. by
PM, elaborates on what that means
Mar 05 DTI/HMT: ‘Economics Paper 11’ &
July 05 HMT: ‘Supporting Growth &
Innovation’
... commits Gov. to driving UK R&D
Spend to 2.5% of GDP (from 1.9%)
... When the Chancellor of The Exchequer
and The Science Minister are aligned
that presents an opportunity
2
So what is Innovation ?
DTI defines Innovation as the successful
exploitation of new ideas
It often involves new technologies or the
application of new technologies
But not always ...
... Many (even most) of the Most-Innovative
Products involve deployment of widely
known technology in Innovative Ways!
3
Innovative Product ... The iPOD
Uses Known technology
Sound compression technology (mp3)
ARM based SoC technology
Operating System / GUI techniques
Micro Hard Disk / SSM
LCD and Touch interfaces
Standard Mobile Phone assembly technology
The Internet (for tune delivery)
The product was a packaging exercise
It displaces Walkman type product
Increasingly displacing home Hi-Fi
... The real innovation however was
iTunes which changes the Dynamic of Music Delivery
4
iTunes ...
iTunes takes Apple into the
PC space
Now bundled loaded on PCs
Creates a new business space
for Apple which to some
degree competes or
collaborates with Microsoft,
Sony & Universal
5
Innovative Product ... The GSM Phone
GSM Telephone
Highly complex technology
Complex but user friendly handsets
Capital intensive infrastructure
... Stimulated move from Analogue to Digital
... Put Nokia ahead of Motorola
... Helped the growth of Vodaphone
and France Telecom
6
GSM Phone ...
Lessons
Standards are important
Open architectures win against closed architectures
GSM provided a better user experience than Analogue
Meanwhile
Qualcomm did well with CDMA
3G coming
New Businesses ARPU
Phone services via the Internet have the opportunity to displace the
Mobile Operators as they once displaced the National Carriers
... The real innovation was creating a Global Standard
7
Innovative Product ... ARM CPU Cell
Uses Known Technology
Was ‘just another’ CPU architecture ...
A good, simple (but basic)
RISC implementation
With good simple (but basic) tools
... MIPS, ARC, 68K and x86 were
around before it
Had no dominating technological
characteristics except that it had
been designed as an ASIC ...
Out of standard cells
On industry standard tools
On industry standard processes
But was small & was cheap
The First ARM ...
~39k transistors
~10 MIP
25mm2 on 1.4um
circa 1991
8
ARM CPU ...
Addressed a market (Embedded Systems) which the ‘big
guys’ saw was too small and specialised
Mobile-Phone Processor.
Replaced a lot of single-chip
1998 80mm , 0.6µ
µm
processor applications
Enabled a new market for SoC
Enabled the concept of the
Virtual Component
Allowed design reuse
Allowed distributed design
AMBA
2
... The First Innovation was to make it
equally Available to Everybody
ARM7TDMI
9
ARM in Embedded Systems Today ...
1990 Acorn ARM Units = 100K
2005 ARM Units 1.7B or 54 every sec
10
The Power of Partnership
open standards, common vision, partner differentiation, local support
11
The Virtual Design Team
Partnership ...
Share the Design Challenge
Allow ‘groups’ to excel in their own areas of expertise
Cost and Return Sharing
Alows ARM’s Partners to focus their R & D for more value add and less
risk
Assures mutual incentive to succeed
More seamless hardware software interfaces
Phasing the cost of Intellectual Property by allowing the customer to
negotiate his up-front payment vs his production payment – License vs
royalty
For the end customer – More choice – Better solutions
... The real ARM Innovation was the shared-cost Licensing and
Royalty Business Model
12
Becoming a Serial Innovator
Then we developed AMBA ...
then Memory Optimised Processors (Thumb),
then Sub-System Assemblies (PrimeXsys),
than Standard OS Ports,
then Drivers for Market Segments (Move, MBX and Jazelle)
then Secure Processors (Secure-Core and Trust-Zone),
then Data-Engines (OptimoDE)
then ...
... Based on ‘standard’ technology, these Innovations delivered
Business advantages ... Increased Confidence for our Users
...They were driven by explicit needs of Real Customers
13
And Some Innovation is Technology
Using ‘standard’ technology to implement the extraordinary.
Developing processors which are optimised for Embedded
Applications
Power management ... IEM
Processors with highest Power Efficiency
Operating Systems
Real Time De-bug
Methods
Cell-libraries ... (Artisan)
TrustZone
... Innovations delivering Technology advantages; enabling the
development of products that could not be done elsewhere
14
So lets look at Innovation
So our retrospective ‘analysis’, tells us that Innovation is a
powerful component of Successful Business ...
But what is it ?
... How do we recognise it ?
... Can we make it happen on
..and..
demand ?
... Can it be tamed by Business
Processes ?
... Some observations from my experience
15
Scaling for Performance
80s & 90s industry growth and scaling in computing power
1982 - Intel 80286
1985 - Intel 80386
1989 - Intel 80386
134 thousand transistors
275 thousand transistors
1.2 million transistors
12MHz; 68.7 mm2
33MHz; 104 mm2
50MHz; 163 mm2
1993 - Intel Pentium
1997 - Intel PentiumII
1999 - Intel PentiumIII
2000 - Intel Pentium4
3.1 million transistors
7.5 million transistors
28 million transistors
42 million transistors
66MHz; 264 mm2
300MHz; 209 mm2
733MHz; 140 mm2
1.5GHz; 224 mm2
16
Scaling for Performance
80s & 90s industry growth and scaling in computing power
1982 - Intel 80286
134 thousand transistors
12MHz; 68.7 mm2
Performance at
expense of power
1985
80386
and- Intel
thermal
275
thousand transistors
challenges
33MHz ; 104 mm2
Move to multiprocessor
1989 - Intel 80386
1.2 million transistors
50MHz; 163 mm2
1993 - Intel Pentium
1997 - Intel PentiumII
1999 - Intel PentiumIII
2000 - Intel Pentium4
3.1 million transistors
7.5 million transistors
28 million transistors
42 million transistors
66MHz; 264 mm2
300MHz; 209 mm2
733MHz; 140 mm2
1.5GHz; 224 mm2
17
Scaling for Power and Area (Cost) Too
Relative size
today!
ARM1 (1986)
3.0µm (2 Layer Metal)
25K Transistors
6MHz
20mW/MHz
50mm2
ARM7TDMI® (2006)
65nm (4 Layer Metal)
100K Transistors
~180MHz
0.025mW/MHz
<0.1mm2
18
Scaling for Power and Area (Cost) Too
Relative size
today!
800 x power
efficiency
550 x area
efficiency
ARM1 (1986)
3.0µm (2 Layer Metal)
25K Transistors
6MHz
20mW/MHz
50mm2
ARM7TDMI® (2006)
65nm (4 Layer Metal)
100K Transistors
~180MHz
0.025mW/MHz
<0.1mm2
19
DFY across process generations
Designs architected for
manufacturability and yield
Working to shorten foundry
data feedback path
45nm
Rigorous DFM/Y methodology
(metrics, qualification)
65nm
Design for variability default for
SRAM (power and performance)
DFM/Y rules adopted as default
Variability and yield analysis
explicit in design margin process
90nm
130nm
Repairable memory
DFM/Y recommendations in DRC
Variability considered for analog
circuits (e.g. sense amp)
2000
2002
2004
2006
20
DFx – where does it fit?
Physical IP provides link between silicon & design
-- and DFx makes it all possible
Specialization
Silicon =
GDS II +
SDF
SoC Design
Silicon =
physics +
chemistry
Physical IP: Digital
EDA
Embedded CPU, cores
Analog
DFx – Design for Mfg, Yield, Test
Manufacturing
21
DFT for soft core IP
IP provider’s objectives:
To ease core integration into an SoC
To enable higher quality of test
DFT methodology:
Wrappers for test isolation
Standardisation – IEEE 1500 and
1450.6
Working with EDA vendors on
reference methodology flows
Provide extra types of test: transition
delay, path delay, IDDQ
Standardised deliverables
SRAM
ROM
MPEG
Blue
Tooth
USB
LCD
GSM
DSP
CPU
22
DFT – the future
Support for 45nm and below
Bridging faults
New memory algorithms
New classes of faults
Devices may have to be shipped with some level of defects
But what level will be acceptable?
Wear-out mechanisms; high energy particles
Need for in-life testing
How to keep cost of test (and repair) at reasonable levels?
Support for DVS (Dynamic Voltage Scaling) systems
Larger SoCs – continue with standardisation effort to make the
DFT flow as smooth as possible
23
What is DFY?
Design for Manufacturability
Things that need to be done to
achieve working silicon
Design for Yield
Things that can be done to increase
amount of working silicon
Examples of DFY
Layout design rules (e.g. line spacing)
Mask fixes (e.g. optical proximity
correction)
Design margining (e.g. Monte Carlo
simulation)
Defect avoidance (e.g. critical area
optimization)
Plus lots more…
24
Components of Yield
Systemic problems
Parametric variations
Defect-related yield
Design-related yield
FOUNDRY
Test-related yield
IP VENDOR
DESIGN TEAM
TEST VENDOR
DFY concerned with all of these
Need to be able to compare effects to enable good tradeoffs
25
What data is needed?
Foundries/Test Houses:
Relative fail rates of different tests (logic, memory, I/O)
Diagnostic data, increasingly at gate level
FOUNDRY
Design team (chip, core)
Critical timing paths, cell usage/criticality statistics
DESIGN
Physical IP vendor
DFM conformance level by cell
Intra-cell path criticality (e.g. memory, complex standard cell)
SRAM data
Test chip organizations
Failure rates for various structures
Challenge: Putting it all together
TEST
TEAM
IP
VENDOR
CHIP
ORGN
26
Practical DFx
Practical (=good) DFx requires only a good methodology and
adequate data
Example: Same layout optimised by several OPC methods –
same areas highlighted as hotspots each time
So we can solve the DFx problem through reasonable
communication between parts of the connected community –
without anyone revealing their crown jewels
27
Current status
ARM is focusing on what is needed to get working silicon
ARM always looking to do more to optimize yield,
manufacturability
Things ARM can do alone
Follow foundry guidelines and establish good layout practices
Incorporate design margin and variability tolerant design
Build yield optimized cells (via doubling, extra spacing)
Develop metrics based on above
Things ARM needs to work with partners to achieve
Anything requiring fab process details (OPC, metal etch, exposure,
etc.)
Anything requiring huge numbers of silicon measurements
variability statistics, modified SPICE models
Precisely quantify yield or yield improvements
Fix systematic foundry yield issues
28
Sustaining Competitive Advantage
An Innovation is not forever ...
Innovations are seldom Patentable
Innovations de-value very quickly
as others follows your lead
A Healthy Company needs a Regular
Influx of Innovations
... Sometimes you have to discard what
has served you well ... and commit to
the new and the unknown !
... “Creative Destruction”
29
Evolving IP Business Model
Connected Community
Development Tools
Software IP
memory
System Level IP:
Processors
Data Engines
SoC
Fabric
Physical IP
30
A Culture for Innovation
New project approval / proposals
You’re not working hard enough
unless you’re making mistakes
Thinking time
– Refresh after burn-out
Capacity for follow-up
Contemplating “Creative
Destruction” ...
Stock markets are Risk Averse
It gets harder the bigger you are
Competitions – e.g. Imperial
31
Where / How Does Innovation Occur
Good University Research points the way to the future
History can help you find some rules, cycles
People who are networked and have more free time are likely
to spot opportunity
Use and correlate all your
sources of information
Spot the growth opportunity
before your competitors
Be prepared to fail but not
destroy the company
Plan and Iterate
– Learn by doing
Read and Learn
32
Applications and services
Messaging
PIM (diary, contacts)
Entertainment
Audio, games
Information services
M-Commerce
DRM
Security
Enterprise
Longer term
Health monitoring / care
33
World of 2015 and Beyond
Digitization and non-intrusive technology creates new markets for
semiconductor products
Trends
Consumer products have shorter shelf life
Increasing Complexity
Standard products
Design for re-use
High volumes needed to recover development cost
Semiconductor company challenges
Ramp-to-volume, system design, software complexity
34
2020: Products for the Infirmed
Clothing measures skin temperature, controls insulation
Mini Logger
Temp monitor
Chilli Heated Clothing
Active wrist, elbow, knee joints to increase strength
Active Joint Brace wins MIT $50K
Implanted automatic drug delivery
Insulin Delivery
Medtronic
Replacement Pancreas
35
2020: Products for the Infirmed
Biometrics - time to visit the doctor again
Controls Parkinson’s
Disease
Replacement of senses
Heart monitoring
Cochlear Ear
Advanced Bionics
36
2020: Products for Kids
Active clothing / skin? that changes colour / tone for mood?
“In his latest column for Business 2.0, "Wearable Tech,"
Rafe Needleman tells us that clothes that can change
colors electronically are soon coming to our closets.”
DNA Screen
On offer in a
Cambridge shop
Eye mounted displays for view ahead with maps, directions
Allow online
searching by thought
37
2020: Products for Kids
Five-sense Virtual Reality gaming and experiences
Holographic projection conference calling with friends
3D freescan image
POC 3D display
38
Summary
Business is always challenging
Great teams can do amazing things
However a company is as weak as its “weakest link”
Opportunity comes from fitting technology to explicit customer needs
Services & applications are as important as product
Open standards win against proprietary standards
Plan and iterate (learning by doing)
Be realistic about people and competition
Have a Global Picture / Vision that is scaleable
Think beyond the possible and back off to reality
We have exciting new challenges but also great opportunities
Have Fun!
39